Multi Agent Supervisor w/ Databricks App

[mmoise]

Hello, how can I create a databricks app using multi agent supervisor (knowledge assistant + genie) with an interface?

[pavannaidu · Accepted solution]

@mmoise 

In the UI, when you create a Databricks App, you can select an existing chat template (see below), which will ask you for your model serving endpoint. In this context, you would provide the Multi Agent Supervisor's endpoint, which will be configured for the Chat App.  

You can use this for reference to build a chat app. There are several templates to get started: https://docs.databricks.com/aws/en/generative-ai/agent-framework/chat-app

I hope this helps. 

[mmoise]

Hi @pavannaidu , how do I improve the speed of the application? When I ask a question it takes a long time to respond.

[pavannaidu]

@mmoise Multi Agent Supervisor (MAS) is an orchestrator agent, so it depends on: 1) how many agents it orchestrates, 2) how complex those agents are, and 3) how complex the user prompt is. 

I would encourage you to review the MLflow experiment traces for the MAS agent to identify bottlenecks and determine how you can optimize the connected agents or provide additional context for MAS to correctly orchestrate. I have seen the biggest improvement when I provide sufficient context for MAS to leverage individual agents.

Lastly, provide examples in the "Examples" and provide guidelines for MAS to execute. This should improve quality and speed. Reference: https://docs.databricks.com/aws/en/generative-ai/agent-bricks/multi-agent-supervisor#step-4-improve-...